Confidential Compute Open Network (COCOON) provides a decentralized platform for running AI inference within Trusted Execution Environments (TEEs) on the TON blockchain. It targets GPU owners seeking to monetize compute power and developers needing secure, private AI inference, ultimately enabling users to access AI services with enhanced confidentiality.

How It Works

COCOON utilizes TEEs, specifically mentioning Intel TDX (Trust Domain Extensions), to create isolated, encrypted environments for AI model execution. This ensures that even the host system cannot access the data or the model during inference. The system operates as a decentralized network where compute providers (workers) serve AI models, and developers or users consume these services, with transactions and coordination facilitated by the TON cryptocurrency.

Quick Start & Requirements

• Worker Setup: Download the latest worker release (ready-to-run TDX image and setup scripts). Setup instructions are included in the release archive.
• Reproducible Build (Optional):
  1. Build VM image: ./scripts/build-image prod
  2. Generate distribution: ./scripts/prepare-worker-dist ../cocoon-worker-dist
  3. Verify TDX image against published checksums.
• Model Building: Use ./scripts/build-model <ModelName> to generate verifiable model archives.
• Prerequisites: Requires hardware supporting Intel TDX. Specific setup scripts and TDX images are provided.

Highlighted Details

• Reproducible Builds: Both worker distributions and model archives can be rebuilt from source, allowing independent verification against published SHA256 checksums.
• Decentralized Compute: Enables GPU owners to earn TON cryptocurrency by providing secure AI inference services.
• Confidential AI: Guarantees privacy and confidentiality for AI computations through TEEs.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or roadmap were provided in the README snippet.

Licensing & Compatibility

The license is stated to be in the LICENSE file; the specific type and any compatibility restrictions for commercial use are not detailed in the provided text.

Limitations & Caveats

The system relies on specific hardware support for Intel TDX, potentially limiting adoption. The setup process involves building VM images and model distributions, which may require significant technical expertise and resources. The project appears to be in a preview stage, suggesting potential for instability or breaking changes.